chore: add Makefile-driven dev toolchain

[wkentaro]

Summary

• Add a Makefile with help/setup/format/lint/test/coverage targets, matching the convention used across the sibling repos so CI and local development share one source of truth.
• lint/format now cover ruff and ty plus taplo (toml), mdformat+mdformat-gfm (md), yamlfix (yaml), and typos; test runs under pytest-xdist.
• .mdformat.toml keeps consecutive ordered-list numbering (number = true); _typos.toml allowlists the OctoMap domain terms leafs (upstream C++ API spelling) and ot (the .ot full-map file extension).
• ci.yml now delegates to make lint (run once) and make test (kept on the 3.10/3.14 matrix).

A few mechanical churn notes for the reviewer:

• taplo reformats pyproject.toml (2-space indent, sorted keys/arrays). Large visual diff but idempotent; comments stay attached.
• mdformat adds blank-after-heading lines in CHANGELOG.md; docs/agents/triage-labels.md's hand-aligned table is compacted by mdformat-gfm.
• yamlfix rewrites the workflow YAML to its canonical style. No behavior change.

Test plan

• make lint (ruff, ty, taplo, mdformat, yamlfix, typos all green)
• make test (40 passed)

[wkentaro]

This was generated by AI during PR processing.

Verdict: recommend-merge — CI fully green (lint, test 3.10/3.14, sdist x2, wheels on all 4 platforms), merge state CLEAN, branch up to date with main (0 commits behind merge-base). History is a clean 3-commit sequence (toolchain → reformat → CI wiring). Reviewed via /code-review (medium); no correctness defects found.

Two non-blocking observations for your call, neither gating the merge:

• CI no longer enforces lockfile freshness. The old ci.yml ran uv sync --locked; the new make lint/make test use plain uv run, which auto-syncs but does not error on an out-of-date uv.lock. A future PR that edits pyproject.toml without refreshing uv.lock would pass CI silently. Add --locked/--frozen to the make targets if you want that guard back.
• ruff.lint.select = ["E", "F", "W"] omits I/UP/ANN and force-single-line from your documented conventions. src is excluded from linting so the blast radius is small; flagging only in case it was unintentional rather than a deliberate minimal set for this PR.